BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

LiveBench Instruction Following

Instruction-following category of LiveBench 1.0, frozen at its final release (25 June 2026). LiveBench 2.0 dropped it.

As of 24 Oct 2026, Gemini 3.8 Flash leads LiveBench Instruction Following on BenchLeader with 81.4%, ahead of Gemini 3.7 Flash at 79.9%, across 67 model configurations with a published result.

Published by
LiveBench
Category
Instruction following
Index weight
Reference only
Models
67
Data as of
24 Oct 2026
Newest result
25 Jun 2026

The publisher has posted no new results on this board since 25 Jun 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.

LiveBench (livebench.ai), White et al. 2024.

What the test looks like

LiveBench's instruction-following tasks: rewriting or summarising recent articles under explicit constraints on length, format and content.

How it is scored

Percent of constraints satisfied, checked automatically, from LiveBench 1.0's final release (25 June 2026). LiveBench 2.0 dropped the category, so these scores no longer update.

What to keep in mind

It checks that rules were followed, not that the writing was good. Models tuned for chattiness lose points for adding unrequested material.

Top 15
  1. 1Gemini 3.8 Flash (high)81.4%
  2. 2Gemini 3.7 Flash (high)79.9%
  3. 3Gemini 3.1 Pro (high)79.1%
  4. 4Muse Spark 1.3 (xhigh)78.0%
  5. 5Qwen3.8-Flash-Next77.1%
  6. 6Claude Fable 5 (max)75.8%
  7. 7Gemini 3.5 Flash (high)75.6%
  8. 8GPT-6 Astra (max)75.6%
  9. 9Gemini 3.6 Flash (high)75.4%
  10. 10Grok 4.7 (xhigh)75.3%
  11. 11Muse Spark 1.2 (xhigh)74.3%
  12. 12GPT-6.1 Sol (max)74.2%
  13. 13Qwen3.8 Max (max)74.1%
  14. 14Qwen3.7 (max)74.0%
  15. 15Smaug Mini73.9%
67 of 67
#
1Gemini 3.8 Flashhigh81.4%64.42026-06-25
2Gemini 3.7 Flashhigh79.9%63.52026-06-25
3Gemini 3.1 Prohigh79.1%60.32026-06-25
4Muse Spark 1.3xhigh78.0%64.92026-06-25
5Qwen3.8-Flash-Next77.1%60.22026-06-25
6Claude Fable 5max75.8%69.32026-06-25
7Gemini 3.5 Flashhigh75.6%63.12026-06-25
8GPT-6 Astramax75.6%70.42026-06-25
9Gemini 3.6 Flashhigh75.4%62.02026-06-25
10Grok 4.7xhigh75.3%61.62026-06-25
11Muse Spark 1.2xhigh74.3%63.22026-06-25
12GPT-6.1 Solmax74.2%69.92026-06-25
13Qwen3.8 Maxmax74.1%63.32026-06-25
14Qwen3.7max74.0%60.62026-06-25
15Smaug Mini73.9%–2026-06-25
16Nemotron 3 Ultra 550B A55B73.5%55.52026-06-25
17Muse Wondermax73.0%67.52026-06-25
18Claude Fable 5.1max73.0%70.52026-06-25
19Qwen3.8 27B72.7%54.62026-06-25
20Claude Opus 4.8max72.0%63.32026-06-25
21Grok 4.671.9%57.82026-06-25
22GPT-5.6 Solmax71.8%68.02026-06-25
23Grok 4.571.5%59.02026-06-25
24Kimi K371.4%59.12026-06-25
25GPT-6.1 Solxhigh71.3%67.72026-06-25
26Deepseek v4 Flash Vision71.0%–2026-06-25
27GPT-5.5xhigh70.7%66.12026-06-25
28Claude Sonnet 5.5xhigh70.5%67.92026-06-25
29GPT-5.4xhigh70.2%64.72026-06-25
30DeepSeek V4.1 Flashmax70.0%61.22026-06-25
31Muse Spark 1.1xhigh69.6%61.62026-06-25
32GLM 5.369.3%–2026-06-25
33Smaug Flash69.3%–2026-06-25
34GPT-6 Solmax68.6%66.42026-06-25
35DeepSeek V4 Pro67.7%53.52026-06-25
36Gemini 3.5 Flash Litehigh67.2%47.92026-06-25
37GPT-5.4 Nanoxhigh67.2%54.22026-06-25
38Claude Opus 5.5xhigh67.0%70.02026-06-25
39Claude Opus 4.7xhigh66.7%57.42026-06-25
40Claude Haiku 5.5xhigh66.5%61.52026-06-25
41GPT-5.2-Codex66.5%57.32026-06-25
42Claude Opus 5.5max65.7%72.62026-06-25
43DeepSeek V4 Flash65.5%52.22026-06-25
44Grok Build 0.165.2%54.82026-06-25
45Mistral Large 4high64.8%49.82026-06-25
46GPT-5.6 Terramax64.6%64.42026-06-25
47Kimi K2.6thinking64.4%59.42026-06-25
48Claude Sonnet 5xhigh63.9%59.02026-06-25
49Claude Opus 5max63.8%68.82026-06-25
50Claude Opus 4.6high63.3%62.72026-06-25
51Claude Sonnet 4.6medium63.2%–2026-06-25
52Grok 4.362.8%54.72026-06-25
53Claude Opus 4.5high62.5%59.22026-06-25
54GLM 5.262.3%52.02026-06-25
55GPT-5.2high61.8%56.02026-06-25
56Ox Alphamax60.3%–2026-06-25
57GPT-5.6 Lunamax60.1%59.92026-06-25
58GPT-5.4 Minixhigh59.8%56.32026-06-25
59Qwen3.6 Plus58.3%58.82026-06-25
60MiniMax M357.5%56.42026-06-25

More instruction following benchmarks · All LLM benchmarks

Cite as: BenchLeader, “LiveBench Instruction Following leaderboard”, data as of 24 Oct 2026, https://www.benchleader.com/benchmarks/livebench_if.

LiveBench Instruction Following: questions

What does LiveBench Instruction Following measure?
LiveBench's instruction-following tasks: rewriting or summarising recent articles under explicit constraints on length, format and content. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads LiveBench Instruction Following?
Gemini 3.8 Flash leads LiveBench Instruction Following with 81.4% as of 24 Oct 2026, ahead of Gemini 3.7 Flash at 79.9%.
How many models have LiveBench Instruction Following results?
67 model configurations have a LiveBench Instruction Following result on BenchLeader, all taken from LiveBench.
Who runs LiveBench Instruction Following and how often is it updated?
LiveBench Instruction Following is published by LiveBench. BenchLeader re-reads the published results once a day and records the date each result was published.
Does LiveBench Instruction Following count toward the BenchLeader Index?
No. LiveBench Instruction Following is shown for reference but left out of the composite index.